On deploy neutron agents should be started before resources creation

[Oleg Bondarev]

Currently on deploy neutron resources are created before neutron agents are started which leads to lots of warnings in logs like:

  WARNING neutron.api.rpc.agentnotifiers.dhcp_rpc_agent_api [req-bb033cc5-b85d-4f02-af43-daa2fa049b8a ea06da003c3f41caa6845536b79921bc e7e7c67a643b4bcaaeae5dddb3a0470c - - -] Unable to schedule network 24b570fc-47d7-44ee-93d3-d7483a7761b2: no agents available; will retry on subsequent port and subnet creation events.

 WARNING neutron.scheduler.l3_agent_scheduler [req-1933ac84-516b-4f5d-ab18-e6e277fea179 ea06da003c3f41caa6845536b79921bc e7e7c67a643b4bcaaeae5dddb3a0470c - - -] No active L3 agents

 WARNING neutron.scheduler.l3_agent_scheduler [req-1933ac84-516b-4f5d-ab18-e6e277fea179 ea06da003c3f41caa6845536b79921bc e7e7c67a643b4bcaaeae5dddb3a0470c - - -] No active L3 agents

and resources are not actually created anywhere except DB.

Though when agents appear resources are scheduled on them it's more a fallback, not a normal flow.
We should start agents first, then create neutron resources.

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: master
Review: https://review.openstack.org/286076

[Sergey Vasilenko (xenolog)]

I disagree, with this sentence:

> Though when agents appear resources are scheduled on them it's more
> a fallback, not a normal flow.
> We should start agents first, then create neutron resources.

Networks and routers is a just record into DB.
Creating without running agents is a ordinary use-case.
Neutron should just re-schedule corresponded resources to the corresponded routers.

Solution, proposed here, looks like workaround. It's not a fix.

[OpenStack Infra (hudson-openstack)]

Change abandoned by Sergey Kolekonov (<email address hidden>) on branch: master
Review: https://review.openstack.org/286076
Reason: The bug is marked as "Won't fix"

[Oleg Bondarev (obondarev)]

@xenolog, I disagree with:
>>Networks and routers is a just record into DB.

They are not just DB records. They are also actual resources which should be applied by agents. Router has no value untill it is processed by l3 agent and applied on the corresponding node.
When someone creates router he/she expects it to be functioning, don't you agree?
For Neutron it's not normal when it is requested to create resource but has no active agents to actually apply it -> so we have warnings in logs which is not good.

I double my question from the review here: Don't you agree that it's logically correct to fully start service (neutron with all agents) first, and then create resources using that service? Can you clarify your resistance? Does this change breaks anything?